Doubao Statistics 2026
By Axis Intelligence Research
Co-author: Sarah Mitchell | Last updated: August 7, 2026 | License: CC BY 4.0
Doubao served 382 million monthly active users in China as of May 2026 and its underlying model family processed more than 180 trillion tokens per day by June 2026 — a ratio of roughly 471,000 tokens per monthly consumer user per day, the widest gap between enterprise token throughput and consumer reach recorded for any Chinese AI platform.
Quick Answer
Doubao (豆包), ByteDance’s consumer AI assistant, reached 382 million monthly active users in China as of May 2026 — 76.6% of the country’s entire AI-native app audience of 499 million, per QuestMobile’s 2026 first-half report. The Doubao model family behind it processed more than 180 trillion tokens per day as of June 2026, and Volcano Engine, ByteDance’s cloud arm, held 49.5% of China’s public-cloud model-as-a-service market by token volume for full-year 2025 per IDC. According to Axis Intelligence Research, the resulting Doubao Token Leverage Index (DTLI™) reads 471,204 as of June 2026, up 146.6% from its September 2025 baseline of 191,083 — meaning ByteDance’s token business is now compounding roughly two and a half times faster than the consumer app that made it famous.
Key Findings
- According to Axis Intelligence Research, Doubao’s 382 million monthly active users in May 2026 represent 76.6% of China’s 499 million AI-native app users — a concentration ratio no other national AI market approaches.
- According to Axis Intelligence Research, the Doubao Token Leverage Index (DTLI™) rose from 191,083 tokens per monthly active user per day in September 2025 to 471,204 in June 2026, a 146.6% increase in ten months.
- Doubao’s daily token call volume passed 180 trillion in June 2026, growing more than tenfold in one year and more than 1,500-fold in two years, according to figures Volcano Engine disclosed at its FORCE conference on June 23, 2026.
- Volcano Engine held 49.5% of China’s public-cloud MaaS market by token volume in full-year 2025, up from 49.2% in the first half, in a market that expanded 16-fold year over year to 1,944 trillion tokens, per IDC.
- According to Axis Intelligence Research, converting just 1% of Doubao’s 382 million monthly active users to its entry 68-yuan subscription tier would generate approximately 3.12 billion yuan (about $459 million) in annualized consumer subscription revenue.
How Many Users Does Doubao Have in 2026?
Doubao is the most-used AI assistant in China by a margin that has widened, not narrowed, through 2026. QuestMobile’s 2026 first-half AI application report, released July 14, 2026, put Doubao at 382 million monthly active users as of May 2026 against 167 million for Alibaba’s Qwen (Qianwen) and 130 million for DeepSeek.
Doubao monthly active users, 2025–2026
| Period | Doubao MAU | Rank in China | Source |
|---|---|---|---|
| August 2025 | 157 million | 1 | QuestMobile via SCMP |
| March 2026 (Q1) | 345 million | 1 | QuestMobile |
| May 2026 | 382 million | 1 | QuestMobile via TechNode |
| June 2026 | 382 million | 1 | QuestMobile via 36Kr |
Doubao’s year-over-year MAU growth to May 2026 was 172.1%, per QuestMobile. That is a striking number for an app that was already the market leader a year earlier — the base effect should have flattened it. It did not, because the growth did not come from the same place twice: the Spring Festival 2026 promotional cycle, expansion into older and lower-tier-city cohorts, and the arrival of agent features all stacked on top of an existing lead.
Daily engagement crossed its own threshold earlier. Doubao’s daily active users exceeded 100 million by late December 2025, a milestone 36Kr reported ByteDance reached with the lowest marketing spend of any company product ever to cross that line.
The 178 million reading, and why we flag it
QuestMobile’s H1 2026 deck reports that Doubao hit an all-time-high reading of 178 million the day after its paid professional edition launched. The report’s English rendering labels that figure as MAU, which cannot be reconciled with the same report’s own 382 million monthly figure. Axis Intelligence Research treats 178 million as a single-day active-user peak and records the labeling discrepancy rather than propagating it. Any publication citing “178 million Doubao MAU” is citing a translation artifact.
Sarah Mitchell, AI/ML: The MAU race in China is the least interesting number in this dataset, and it is the only one most coverage reports. Doubao winning a user-count contest against Qwen and DeepSeek was settled eighteen months ago. What changed in 2026 is that ByteDance stopped optimizing the app for reach and started optimizing it for session depth and paid conversion — and the session-length data shows the transition is still incomplete.
How Deep Is Doubao Engagement Compared to DeepSeek and Kimi?
Reach and depth are separate businesses, and Doubao is winning one of them decisively while placing third in the other.
| App | Sessions over 10 minutes (June 2026) | YoY change | Source |
|---|---|---|---|
| DeepSeek | 30.0% | +7.8 pp | QuestMobile |
| Doubao | 27.5% | +2.2 pp | QuestMobile |
| Kimi | 26.1% | +8.4 pp | QuestMobile |
QuestMobile’s full H1 2026 deck also reports that 40.6% of Doubao sessions run under three minutes — the profile of a fragmented, high-frequency utility rather than a deep-work tool. DeepSeek sits at the opposite pole, positioned around long-form reasoning.
Category-level engagement, for context: AI-native apps in China averaged 92.7 uses and 183.0 minutes per user per month as of May 2026, up 58.3% and 40.0% year over year respectively. That is the highest stickiness of any app category QuestMobile tracks, ahead of AI plugins (60.9 uses), device-maker AI assistants (51.4), PC clients (26.9) and PC web (25.3).
The displacement is measurable elsewhere. Traditional search in China recorded a 19.1% year-over-year decline in per-user frequency and a 13.5% decline in duration in May 2026. Across all app categories QuestMobile tracks, 28% fell on both frequency and duration year over year, and 40% fell on one. Our chatbot statistics analysis tracks the equivalent substitution pattern in Western markets.
Sarah Mitchell: Doubao’s +2.2 points on deep sessions against Kimi’s +8.4 is the tell. Kimi is growing depth from a small base because that is the only game it can win. Doubao is growing depth slowly because 382 million users includes an enormous tail of people asking it one question a day. Both trajectories are rational. Only one of them supports a 500-yuan subscription tier.
How Many Tokens Does Doubao Process Per Day?
This is the number that actually explains ByteDance’s AI position, and it is reported far less often than the user count.
| Date | Daily tokens | Growth reference | Source |
|---|---|---|---|
| May 2024 (launch) | ~120 billion | Baseline | Volcano Engine |
| September 2025 | 30 trillion | 253× vs May 2024 | Volcano Engine |
| December 2025 | 50 trillion | 417× vs May 2024 | Volcano Engine |
| April 2, 2026 | 120 trillion | ~1,000× vs May 2024 | Volcano Engine via China Daily |
| June 2026 | 180 trillion | >1,500× vs May 2024 | Volcano Engine (FORCE 2026) |
Volcano Engine confirmed the 120-trillion figure on April 2, 2026, noting the volume had doubled in the preceding three months. Ten weeks later, at the FORCE conference on June 23, 2026, president Tan Dai put the figure above 180 trillion.
Enterprise concentration
| Metric | Value | Date | Source |
|---|---|---|---|
| Enterprises consuming over 1T tokens cumulatively | 100 | Dec 2025 | Volcano Engine |
| Same measure | 140 | April 2026 | Volcano Engine |
| “Trillion Token Club” (annual call volume over 1T) | 200 | June 2026 | Volcano Engine |
| Enterprises and individuals using Volcano Ark | over 1.1 million | June 2026 | Volcano Engine |
The heavy-user cohort doubled inside six months. That is the shape of an infrastructure business finding its anchor tenants, not a consumer app scaling.
The Doubao Token Leverage Index (DTLI™)
What it measures. DTLI™ — the Doubao Token Leverage Index — expresses how much model throughput ByteDance’s ecosystem generates for every consumer user of the Doubao app. It is the single ratio that separates ByteDance’s AI position from every competitor that publishes only a user count.
Formula.
DTLI™ = Daily tokens served by the Doubao model family ÷ Doubao consumer app monthly active users
Inputs and readings.
| Reading date | Daily tokens | Doubao MAU (matched period) | DTLI™ | Change vs baseline |
|---|---|---|---|---|
| September 2025 (baseline) | 30,000,000,000,000 | 157,000,000 (Aug 2025) | 191,083 | — |
| April 2026 | 120,000,000,000,000 | 345,000,000 (Mar 2026) | 347,826 | +82.0% |
| June 2026 | 180,000,000,000,000 | 382,000,000 (Jun 2026) | 471,204 | +146.6% |
Worked example for the current reading: 180,000,000,000,000 ÷ 382,000,000 = 471,204 tokens per monthly active user per day.
How to read it. DTLI™ is deliberately a cross-population ratio. The numerator is platform-wide — it includes enterprise API traffic on Volcano Ark, ByteDance’s internal Douyin and Feishu workloads, Seedance video generation and Seedream image generation. The denominator is the consumer app alone. The index therefore does not describe what an individual Doubao user consumes. It describes how far the token economy behind the brand has outrun the brand’s consumer surface.
A rising DTLI™ means enterprise, agent and multimodal workloads are compounding faster than consumer reach. A flat DTLI™ would mean the app is carrying the growth. Between September 2025 and June 2026 the index rose 146.6% while MAU rose 143.3% — closer than the headline suggests, and the honest reading is that both engines fired at once, with tokens marginally ahead. What breaks the tie is composition: the enterprise cohort above one trillion tokens doubled in six months while consumer MAU grew by 37 million in three.
Scope of the reading. DTLI™ is a structural indicator, not a per-user consumption estimate. Both series are vendor-disclosed (tokens) or panel-measured (MAU); neither is audited. Token disclosures are point-in-time announcements at conferences rather than a continuous series, so readings are anchored to disclosure dates and matched to the nearest published MAU. This is the baseline series; subsequent readings will be added to this page as Volcano Engine discloses new token figures.
Methodology version: DTLI 1.0, published August 7, 2026. Licensed CC BY 4.0. Citation: Axis Intelligence Research, Doubao Token Leverage Index (DTLI™), 2026.
What Is Doubao’s Market Share in China’s AI Market?
Two different market-share questions get conflated constantly. They have different answers.
Consumer app share
| App | MAU (May 2026) | Share of 499M AI-native total | Source |
|---|---|---|---|
| Doubao | 382 million | 76.6% | QuestMobile |
| Qwen (Qianwen) | 167 million | 33.5% | QuestMobile |
| DeepSeek | 130 million | 26.1% | QuestMobile |
Shares exceed 100% because users run multiple apps; QuestMobile counts each app’s MAU independently. According to Axis Intelligence Research, Doubao’s 76.6% penetration of China’s AI-native app base is the highest single-app concentration in any major national AI market.
Enterprise MaaS share
| Vendor | Share of China public-cloud MaaS by tokens | Period | Source |
|---|---|---|---|
| Volcano Engine | 49.5% | FY2025 | IDC |
| Alibaba Cloud | 27% | H1 2025 | IDC |
| Baidu AI Cloud | 17% | H1 2025 | IDC |
The market itself grew 16-fold year over year to 1,944 trillion tokens in 2025, from 114.2 trillion in 2024. Volcano Engine’s share rose from 49.2% in H1 2025 to 49.5% for the full year — expansion during the most competitive stretch, when nearly every Chinese cloud vendor entered the segment.
The methodology caveat that matters. IDC’s MaaS measurement covers external enterprise calls to models on public clouds. It excludes ByteDance’s own AI applications — Doubao and Jimeng among them — and excludes tokens generated when Douyin and Feishu deploy models internally. So the 49.5% share and the 180 trillion daily tokens are not the same universe, and any analysis that divides one by the other produces a meaningless number. That exclusion is why DTLI™ is framed as a leverage ratio rather than a share calculation.
Sarah Mitchell: Volcano Engine’s 49.5% is usually explained as the residue of the 2024 price war, when it cut Doubao Pro-32K input pricing to 0.0008 yuan per thousand tokens — roughly a 99.3% discount to the prevailing rate. That explanation stopped working in 2025, when every competitor matched the price and the share still went up. Prefill-decode disaggregation and KV caching only pay for themselves above a certain call volume; below it, the cache infrastructure costs more than the compute it saves. The moat is the scale, and the price was the mechanism for buying it.
What Does Doubao Cost? Subscription and API Pricing
Doubao spent its first three years free. That ended in 2026.
Consumer subscription tiers
| Tier | Monthly price (CNY) | Approx. USD | Source |
|---|---|---|---|
| Entry | ¥68 | $10 | The Beijing News via Seoul Economic Daily |
| Mid | ¥200 | $29 | The Beijing News via Seoul Economic Daily |
| Top | ¥500 | $74 | The Beijing News via Seoul Economic Daily |
The tiers surfaced on Doubao’s App Store listing on May 4, 2026 and immediately topped Weibo’s hot-search list. The paid professional edition formally launched June 24, 2026, targeting six office scenarios — presentation building, precise data analysis and video production among them — with a synced mobile and PC client. The free tier remains.
API pricing (Volcano Ark)
| Model | Input (per M tokens) | Output (per M tokens) | Cache hit | Source |
|---|---|---|---|---|
| Doubao-Seed-2.1-Pro | ¥6 | ¥30 | ¥1.2 | Volcano Engine, June 23 2026 |
| Doubao-Seed-2.1-Turbo | ¥3 | ¥15 | ¥0.6 | Volcano Engine, June 23 2026 |
| Doubao-Seed-Code (promo) | ¥9.9 first month, then ¥40/month | — | — | Volcano Engine |
Volcano Engine claims Doubao 2.1 Pro’s total cost of ownership runs roughly 80% below the Claude Opus 4.6 series. That is a vendor claim about a competitor’s economics, and it should be read as one. For the cross-vendor picture, see our AI inference cost statistics.
The price-card bracket on 180 trillion tokens
According to Axis Intelligence Research, applying published Doubao 2.1 Pro list rates to a single day’s disclosed throughput brackets the notional value of that traffic between ¥216 million per day (all 180T priced at the ¥1.2 cache-hit floor) and ¥1.08 billion per day (all 180T priced at the ¥6 standard input rate).
180,000,000,000,000 ÷ 1,000,000 = 180,000,000 million-token units × ¥1.2 = ¥216,000,000 · × ¥6 = ¥1,080,000,000
This is a price-card bracket, not a revenue estimate, and the gap between the two figures is the point. A large share of that 180 trillion is ByteDance’s own internal traffic, which carries no external price; the remainder splits across cache hits, Turbo-tier calls and negotiated enterprise rates well below list. IDC has separately noted that Volcano Engine’s revenue share of the MaaS market ranks first but sits a few percentage points below its call-volume share — its per-token realized price is below the industry average. Anyone converting daily tokens into ByteDance AI revenue is producing a number with no defensible basis, and we are not going to be the source they cite for it.
How Does Doubao 2.1 Pro Perform Against Frontier Models?
ByteDance released Doubao-Seed-2.1-Pro on June 23, 2026, positioned around coding, agent execution and vision-language work.
| Benchmark | Capability | Reported result | Source |
|---|---|---|---|
| Terminal Bench | Coding | Roughly on par with Claude Opus 4.7 | Volcano Engine (self-reported) |
| SciCode | Scientific computing | 59.8, above Opus 4.7 and GPT-5.5 | Volcano Engine (self-reported) |
| NL2Repo | Repo-level code generation | 47, above GPT-5.5 and Gemini 3.1 | Volcano Engine (self-reported) |
| MCP Atlas | Agent / tool calling | Above Opus 4.7 and GPT-5.5 | Volcano Engine (self-reported) |
Every figure in that table is vendor-disclosed and competitor scores were not published alongside them. MCP Atlas covers 36 MCP servers, 220 tools and over 1,000 tasks. Model lineage: Doubao 1.6 (June 2025), 1.8 (December 18, 2025), Seed 2.0 with a 256K context window (February 14, 2026), Seed 2.1 (June 23, 2026).
Sarah Mitchell: Self-reported benchmarks with the comparison scores withheld are marketing, and I would say so about any vendor. The disclosure worth more than the whole table is the production anecdote Tan Dai gave: an RTL chip-design task where the model ran nine iterations over roughly eighteen hours and produced six modules and about 1,300 lines of code. Whether that number is exact matters less than what it concedes — that the useful unit of measurement has moved from a benchmark score to a task duration, and eighteen hours of autonomous execution is not something a leaderboard can express.
How Much Could Doubao Earn From Subscriptions?
Doubao’s paid tiers landed in a market that has resisted consumer AI subscriptions. A Tencent Research Institute survey of 3,570 Chinese adults conducted in September 2025 found 96.2% had used generative AI, but only 16.1% had ever paid for it. Baidu’s Ernie Bot ran a paid membership, then made the service entirely free and refunded existing members.
According to Axis Intelligence Research, applying conversion scenarios to Doubao’s 382 million monthly active users at the ¥68 entry tier produces the following annualized ceiling:
| Conversion rate | Paying users | Annual revenue (CNY) | Annual revenue (USD) |
|---|---|---|---|
| 1% | 3.82 million | ¥3.12 billion | $459 million |
| 3% | 11.46 million | ¥9.35 billion | $1.38 billion |
| 5% | 19.10 million | ¥15.59 billion | $2.29 billion |
382,000,000 × 1% = 3,820,000 subscribers 3,820,000 × ¥68 × 12 months = ¥3,117,120,000 ÷ 6.8 CNY/USD = $458.4 million
These are scenario computations at the entry price, not forecasts. ByteDance has disclosed no subscriber count. The conversion basis is USD 1 = CNY 6.8, derived from Volcano Engine’s own published conversions at the June 2026 FORCE conference; higher tiers at ¥200 and ¥500 would scale the figures proportionally.
The wider context is a market-structure reversal. Chinese cloud vendors including Alibaba, Tencent and Baidu raised service prices by up to roughly 30% in the second quarter of 2026, ending a three-year discount cycle. DeepSeek, Zhipu AI and MiniMax have all moved toward paid tiers or prepaid plans, and DeepSeek signalled significant API price increases in August 2026. The driver is agent workloads: a single agent task consumes orders of magnitude more tokens than a chatbot turn, and free-tier economics stop working at that consumption profile.
Where Does Doubao Sit in China’s Overall AI Adoption?
China had 602 million generative AI users as of December 2025, up 141.7% year over year, at 42.8% population penetration, per CNNIC’s 57th Statistical Report on China’s Internet Development, released February 5, 2026. China’s internet population stood at 1.125 billion, an 80.1% penetration rate.
According to Axis Intelligence Research, Doubao’s 382 million monthly active users represent approximately 63.5% of that 602 million generative-AI base — a reach ratio no single Western assistant achieves in its home market. That comparison spans a base-period gap: the CNNIC figure is December 2025 and the Doubao figure is May 2026, so 63.5% should be read as an upper bound that a refreshed CNNIC base will pull downward. We publish it with the mismatch stated rather than silently aligning the dates.
The composition of Chinese AI usage explains why Doubao’s session profile skews short. CNNIC found 76% of Chinese generative-AI users treat these tools primarily for question answering, 47.8% for image or video work, and 10.8% for code generation. That is a consumer utility distribution, and it is closely mirrored in Western usage data. For the full national picture, see our China AI statistics analysis, and for the regulatory frame governing every registered service in that market, our AI regulation by country map.
Sarah Mitchell: Doubao’s competitive position is not really contested inside China. What is contested is whether ByteDance can carry that position across a border. Doubao’s international sibling — launched as Cici, now Dola — has never travelled the way DeepSeek’s weights travelled or the way Qwen travelled through the open-source ecosystem. Doubao is enormous in usage data and nearly invisible in the global developer conversation, and those two facts have coexisted for two years without resolving.
Methodology
Collection. Every figure in this dataset was retrieved from a source document opened during production between August 6 and August 7, 2026. Nothing is reconstructed from prior knowledge. Where an original Chinese-language source was reported through an English intermediary, both the originating organization and the reporting outlet are recorded in the CSV.
Source hierarchy. Volcano Engine and ByteDance disclosures are treated as company-reported. QuestMobile figures are panel measurements, not audited counts. IDC market-share figures are analyst estimates with a stated scope. CNNIC is a government statistical report. Vendor benchmark claims are labeled self-reported in both the article and the CSV.
Currency. USD conversions use CNY 6.8 = USD 1, derived from Volcano Engine’s own published conversions at the June 23, 2026 FORCE conference. Where a source published its own USD figure, that figure is retained.
DTLI™ computation. Daily token disclosures are point-in-time announcements rather than a continuous series. Each DTLI™ reading pairs a disclosed token figure with the nearest published QuestMobile MAU, and both dates appear in the CSV. The September 2025 reading is the baseline.
Scope and caveats. Doubao’s token numerator spans the whole ByteDance model ecosystem, including internal Douyin and Feishu workloads and multimodal generation, while IDC’s market-share denominator explicitly excludes ByteDance first-party applications — the two are not interchangeable and are never divided against one another here. ByteDance has published no subscriber count, revenue figure or ARPU for Doubao; the subscription table is scenario arithmetic at list price, clearly labeled. QuestMobile does not publish confidence intervals for its panel.
What this reading does not capture. Doubao usage embedded in Douyin, Feishu, the Doubao smartphone, Doubao AI glasses and third-party device integrations sits outside QuestMobile’s AI-native app category and is not counted in the 382 million figure. Overseas Doubao-family usage through Cici/Dola is excluded entirely.
About This Dataset
Title: Doubao Statistics 2026 — Users, Tokens, Market Share and Monetization
Creator and publisher: Axis Intelligence Research
File: doubao-statistics-2026.csv
Coverage: May 2024 – June 2026 · China, with global model comparisons
License: CC BY 4.0
Citation formats
APA — Axis Intelligence Research. (2026). Doubao statistics 2026: Users, tokens, market share and monetization. https://axis-intelligence.com/doubao-statistics/
MLA — Axis Intelligence Research. “Doubao Statistics 2026: Users, Tokens, Market Share and Monetization.” Axis Intelligence, 2026, axis-intelligence.com/doubao-statistics/.
Chicago — Axis Intelligence Research. “Doubao Statistics 2026: Users, Tokens, Market Share and Monetization.” Axis Intelligence, 2026. https://axis-intelligence.com/doubao-statistics/.
Frequently Asked Questions
Does Doubao’s 382 million MAU count include Douyin and Feishu AI usage?
No. QuestMobile’s AI-native app category counts the standalone Doubao app only. AI features embedded inside Douyin, Feishu, the Doubao smartphone and device-maker assistants fall into separate categories — device-vendor AI applications alone reached 755 million MAU in May 2026 at much lower usage frequency. Treating those as additive to Doubao’s 382 million double-counts users.
Why do Volcano Engine’s 180 trillion daily tokens and IDC’s 49.5% MaaS share not reconcile?
They measure different universes. IDC’s MaaS market share covers external enterprise calls to models on public clouds and explicitly excludes ByteDance’s own applications — Doubao, Jimeng — and internal Douyin and Feishu deployments. Volcano Engine’s 180 trillion is the full model-family throughput including all of that. Dividing one into the other produces a number with no meaning.
Is Doubao still free in 2026?
Yes, the free tier remains. Doubao added three paid subscription tiers at ¥68, ¥200 and ¥500 per month, surfaced on the App Store May 4, 2026 and formally launched as a professional edition on June 24, 2026. The paid tiers target compute-heavy office work — deck building, data analysis, video generation — rather than gating conversation.
How does Doubao’s per-token API pricing compare to frontier Western models?
Doubao-Seed-2.1-Pro lists at ¥6 per million input tokens and ¥30 per million output, with cache hits at ¥1.2 — roughly $0.88 and $4.42 at 6.8 CNY/USD. Volcano Engine claims a total cost of ownership around 80% below the Claude Opus 4.6 series, a vendor claim about a competitor’s pricing rather than an independent benchmark. Realized enterprise pricing runs below list; IDC notes Volcano Engine’s revenue share sits several points under its call-volume share.
What does the DTLI™ actually tell an analyst that MAU does not?
MAU measures how many people opened an app. DTLI™ measures how much model throughput the surrounding ecosystem generates per unit of that consumer reach. A platform whose DTLI™ climbs while MAU is flat is converting into enterprise infrastructure; one whose DTLI™ falls is a consumer product that has outgrown its backend monetization. Doubao’s index rose 146.6% between September 2025 and June 2026 while MAU rose 143.3% — both engines running, with tokens marginally ahead and the enterprise cohort concentrating faster.
Why does DeepSeek beat Doubao on session depth despite one-third the users?
DeepSeek is positioned around long-chain reasoning and draws a user base that arrives with harder tasks — 30.0% of its June 2026 sessions exceeded ten minutes against Doubao’s 27.5%. Doubao’s distribution runs the other way, with 40.6% of sessions under three minutes. Neither profile is a defect; they reflect deliberate product positioning, and the short-session profile is what makes Doubao’s 382 million reachable in the first place.
Has Doubao’s paid tier reduced engagement?
Not on the evidence available. QuestMobile recorded Doubao’s highest single-day active-user reading, 178 million, the day after the professional edition launched, and the app added 13.78 million new users in June 2026 against May. ByteDance has disclosed no subscriber count, so conversion remains unmeasured — engagement holding is not the same as monetization succeeding.
What would move Volcano Engine’s 49.5% MaaS share?
Three things, in order of likelihood: a competitor achieving comparable inference economics at scale, since prefill-decode disaggregation and KV caching only pay for themselves above high call volumes; a large enterprise cohort standardizing on a rival’s agent harness, which raises switching costs beyond the API layer; or compute supply constraints forcing rationing. Price alone will not do it — competitors already matched Volcano Engine’s 2024 cuts and its share still rose through 2025.